arXiv subjects

Rishi Gupta

Publications and source records attributed to Rishi Gupta.

5 recordsLinked to original sources

arXiv2025

O3SLM: Open Weight, Open Data, and Open Vocabulary Sketch-Language Model

While Large Vision Language Models (LVLMs) are increasingly deployed in real-world applications, their ability to interpret abstract visual inputs remains limited. Specifically, they struggle to comprehend hand-drawn sketches, a modality that offers an intuitive means of expressing concepts that are difficult to describe textually. We identify the primary bottleneck as the absence of a large-scale dataset that jointly models sketches, photorealistic images, and corresponding natural language instructions. To address this, we present two key contributions: (1) a new, large-scale dataset of image-sketch-instruction triplets designed to facilitate both pretraining and instruction tuning, and (2) O3SLM, an LVLM trained on this dataset. Comprehensive evaluations on multiple sketch-based tasks: (a) object localization, (b) counting, (c) image retrieval i.e., (SBIR and fine-grained SBIR), and (d) visual question answering (VQA); while incorporating the three existing sketch datasets, namely QuickDraw!, Sketchy, and Tu Berlin, along with our generated SketchVCL dataset, show that O3SLM achieves state-of-the-art performance, substantially outperforming existing LVLMs in sketch comprehension and reasoning.

cs.CV↗

arXiv2025

Chemist-aligned retrosynthesis by ensembling diverse inductive bias models

Chemical synthesis remains a critical bottleneck in the discovery and manufacture of functional small molecules. AI-based synthesis planning models could be a potential remedy to find effective syntheses, and have made progress in recent years. However, they still struggle with less frequent, yet critical reactions for synthetic strategy, as well as hallucinated, incorrect predictions. This hampers multi-step search algorithms that rely on models, and leads to misalignment with chemists' expectations. Here we propose RetroChimera: a frontier retrosynthesis model, built upon two newly developed components with complementary inductive biases, which we fuse together using a new framework for integrating predictions from multiple sources via a learning-based ensembling strategy. Through experiments across several orders of magnitude in data scale and splitting strategy, we show RetroChimera outperforms all major models by a large margin, demonstrating robustness outside the training data, as well as for the first time the ability to learn from even a very small number of examples per reaction class. Moreover, industrial organic chemists prefer predictions from RetroChimera over the reactions it was trained on in terms of quality, revealing high levels of alignment. Finally, we demonstrate zero-shot transfer to an internal dataset from a major pharmaceutical company, showing robust generalization under distribution shift. With the new dimension that our ensembling framework unlocks, we anticipate further acceleration in the development of even more accurate models.

cs.LG↗

arXiv2016

A PAC Approach to Application-Specific Algorithm Selection

The best algorithm for a computational problem generally depends on the "relevant inputs," a concept that depends on the application domain and often defies formal articulation. While there is a large literature on empirical approaches to selecting the best algorithm for a given application domain, there has been surprisingly little theoretical analysis of the problem. This paper adapts concepts from statistical and online learning theory to reason about application-specific algorithm selection. Our models capture several state-of-the-art empirical and theoretical approaches to the problem, ranging from self-improving algorithms to empirical performance models, and our results identify conditions under which these approaches are guaranteed to perform well. We present one framework that models algorithm selection as a statistical learning problem, and our work here shows that dimension notions from statistical learning theory, historically used to measure the complexity of classes of binary- and real-valued functions, are relevant in a much broader algorithmic context. We also study the online version of the algorithm selection problem, and give possibility and impossibility results for the existence of no-regret learning algorithms.

cs.LG↗

arXiv2014

Decompositions of Triangle-Dense Graphs

High triangle density -- the graph property stating that a constant fraction of two-hop paths belong to a triangle -- is a common signature of social networks. This paper studies triangle-dense graphs from a structural perspective. We prove constructively that significant portions of a triangle-dense graph are contained in a disjoint union of dense, radius 2 subgraphs. This result quantifies the extent to which triangle-dense graphs resemble unions of cliques. We also show that our algorithm recovers planted clusterings in approximation-stable k-median instances.

cs.DS↗

arXiv2012

Compressive Sensing with Local Geometric Features

We propose a framework for compressive sensing of images with local distinguishable objects, such as stars, and apply it to solve a problem in celestial navigation. Specifically, let x be an N-pixel real-valued image, consisting of a small number of local distinguishable objects plus noise. Our goal is to design an m-by-N measurement matrix A with m << N, such that we can recover an approximation to x from the measurements Ax. We construct a matrix A and recovery algorithm with the following properties: (i) if there are k objects, the number of measurements m is O((k log N)/(log k)), undercutting the best known bound of O(k log(N/k)) (ii) the matrix A is very sparse, which is important for hardware implementations of compressive sensing algorithms, and (iii) the recovery algorithm is empirically fast and runs in time polynomial in k and log(N). We also present a comprehensive study of the application of our algorithm to attitude determination, or finding one's orientation in space. Spacecraft typically use cameras to acquire an image of the sky, and then identify stars in the image to compute their orientation. Taking pictures is very expensive for small spacecraft, since camera sensors use a lot of power. Our algorithm optically compresses the image before it reaches the camera's array of pixels, reducing the number of sensors that are required.

cs.CG↗